Data-Driven Approaches to Facilitate Inorganic Materials Discovery and Synthesis

Machine learning is playing an increasingly prominent role in computational materials discovery. Recent advances in (universal) machine learning interatomic potentials allow practitioners to optimize the structure and compute the energy of inorganic crystals for arbitrary material compositions. These computed energetics can then be used as inputs to the convex hull analysis to determine if a hypothetical material is thermodynamically stable against potential competing phases. While stability is an important property for a candidate material, it does not determine if or how that material can be made in the lab. Whereas stability (at some set of conditions) is an intrinsic property of a material, synthesizability is not a similarly straightforward binary – it depends on experimental choices such as precursors, temperature, synthesis approach, etc. This talk will discuss recent efforts to use machine learning in the context of synthesis “recipe” generation, assessment, and optimization.